{"id":"0edf0be8-b137-4cb8-8aaa-e35778d3fb3f","arxiv_id":"2411.14919","paper_version":3,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper derives the optimal multi-user beamforming structure for continuous aperture arrays and provides a globally optimal algorithm plus near-optimal MRT, ZF, and MMSE designs.","lead":"This paper works out the optimal way to beamform from a continuous aperture array, a hypothetical antenna made of a smooth sheet of radiating material. It gives the exact structure of the optimal beamformer, a global search algorithm, and simple close-to-optimal designs, showing gains over conventional discrete antenna arrays.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1 rests on an unproven strong-duality claim for a nonconvex infinite-dimensional problem; the proof's 'infinitesimal vectors' analogy does not establish the required duality.","rationale":"The paper's main deliverable is a closed-form optimal structure and a global algorithm. The proof chain is: Lemma 3 (full power), Theorem 1 (structure), Algorithm 1/2 (global optimization). The structure is derived from KKT conditions of problem (14), and KKT is claimed sufficient because the problem is 'convex and satisfies strong duality'. This is the only point where the infinite-dimensional nature of CAPA is addressed by analogy rather than analysis. The original constraint (14b) is not convex as written, so the proof contains a genuine gap. If that gap cannot be repaired, the optimal-structure theorem may fail or hold only locally, and every downstream result (the MRT/ZF/MMSE optimality claims, the polyblock global optimality, and the numerical benchmark) loses its grounding. The concern is not that the result disagrees with consensus: a same-form result is known for discrete arrays, so the continuous extension is plausible. The issue is that the continuous extension requires compactness and constraint-qualification arguments that are absent. The fixed-point convergence of (27) is a second gap, but it is algorithmic rather than foundational. I agree with the reader's weakest_assumption, and the reader's CONDITIONAL verdict remains appropriate. The concern is concrete and testable, but the analogy to [34] and the numerical demonstrations suggest the result is likely correct; hence the verdict should stay conditional rather than being upgraded or downgraded.","tokens_in":21250,"tokens_out":7289,"duration_ms":75173,"concrete_test":"Independently re-derive Theorem 1: reformulate (14) with phase-normalized variables (e.g., Im∫_S h_k^* w_k = 0, Re∫_S h_k^* w_k ≥ 0) as a second-order cone program in the Hilbert space L2(S), and verify Slater's condition under the Green's-function channel model (49) on bounded S. If a correct SOCP formulation with Slater holds, strong duality follows and (15) is validated; if the phase normalization is not equivalent or Slater fails, produce a concrete channel realization where the stationary point from (73)-(74) is not globally optimal. As a complementary numerical check, discretize S using N=4 Gauss-Legendre points and compare the global optimum of the discretized problem (found by exhaustive or SDP-based global search) with the structure (15) evaluated on the discretized Q; any mismatch would refute the continuous-space claim.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Appendix D opens by asserting that problem (14) 'is convex and satisfies strong duality', justified only by 'considering h_k(s) and w_k(s) as infinitesimal vectors'. This is the load-bearing step for Theorem 1 and for the global-optimality claim of Algorithm 1, but it is not proven. As written, (14b) is not a convex constraint: already for K=1 the feasible set {w_1 : |∫_S h_1^* w_1|^2 ≥ η_1 σ^2} is the complement of a slab in L2(S) and is not convex, so the problem cannot be called convex without an additional phase-normalization or conic reformulation. No such reformulation is supplied for the infinite-dimensional setting, and the 'infinitesimal vectors' analogy with finite-dimensional MIMO does not by itself establish Slater's condition or strong duality in L2(S). If strong duality fails, the KKT stationarity condition (73)-(74) may characterize only a stationary point, and the closed-form structure (15) need not be globally optimal; the polyblock projection in Algorithm 2, based on (26)-(27), would then inherit the error. A secondary gap is that the fixed-point iteration (27) is stated without a convergence or uniqueness proof, so even if (15) is correct, the proposed algorithm's global convergence is not fully demonstrated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper considers downlink multi-user beamforming for continuous aperture array (CAPA) systems, where each beamformer is a continuous function over the aperture. The central contribution is a claimed closed-form optimal beamforming structure, w(s) = h(s)(I_K + σ^{-2} Λ Q)^{-1} P^{1/2}, obtained by combining Lagrangian duality, calculus of variations, and an inversion lemma for continuous functions. Based on this structure, the paper develops a polyblock-based monotonic optimization algorithm for globally optimal beamforming, including a fixed-point iteration for the power-minimization subproblem, and derives low-complexity MRT, ZF, and MMSE designs. Numerical results show that CAPA outperforms spatially discrete arrays and that the MMSE design is nearly optimal in many regimes.","tokens_in":21456,"tokens_out":6644,"duration_ms":69487,"significance":"If the optimality claims are fully established, the paper provides a general structural result for CAPA multi-user beamforming that unifies global optimization and heuristic designs, and it gives concrete insight into the performance gap between continuous and discrete apertures. The derivation is self-contained, the algebraic steps in the appendices are largely correct, and the numerical validation covers multiple operating regimes. The main value is the closed-form structure and the explicit comparison with SPDA designs. However, the proof of global optimality relies on an unproven strong-duality assertion for an infinite-dimensional nonconvex problem, and the proposed fixed-point iteration lacks a convergence proof; these gaps currently limit confidence in the central claims.","major_comments":[{"comment":"The proof of Theorem 1 begins by asserting that problem (14) is convex and satisfies strong duality, justified only by 'considering h_k(s) and w_k(s) as infinitesimal vectors.' This is not established. As written, the feasible set defined by |∫_S h_k^*(s)w_k(s)ds|^2 ≥ η_k(...) is not convex; already for K=1 it is the complement of a slab in L^2(S). A convex reformulation is possible by phase-rotating each w_k so that ∫_S h_k^*(s)w_k(s)ds is real and nonnegative and rewriting the constraints as second-order cone constraints, but this reformulation is not supplied, and Slater's condition or another strong-duality condition in infinite dimensions is not proved. Since KKT stationarity alone is only necessary for the original nonconvex problem, the global optimality of the structure (15) and of Algorithm 1 rests on this missing step.","section":"Appendix D, problem (14)"},{"comment":"The fixed-point iteration λ_k^{(n+1)} = f_k(α, λ^{(n)}) is stated without a proof of convergence or uniqueness. Algorithm 2 relies on this iteration to compute the projection π_G(z) via the condition P^* = Σ_k λ_k ≤ P, and the global optimality of the polyblock method depends on the projection being correct. A convergence proof, for example showing that the iteration is a contraction or monotonically converges to the unique fixed point, is needed before the global optimality claim for Algorithm 1 is complete.","section":"Section IV-B, Eq. (27)"}],"minor_comments":[{"comment":"In the matrix calculation, D is written as (I_K - Λ Q)^{-1}, but D was defined as (I_K + Λ Q)^{-1}. The conclusion α_{k,i}=0 remains correct with the definition (I_K + Λ Q)^{-1}, so this appears to be a sign typo in the displayed derivation.","section":"Lemma 2 proof, Appendix B, Eq. (60)"},{"comment":"The symbol P is overloaded: in Eq. (15) it is a diagonal power-allocation matrix, while in Eq. (16) it is the scalar power budget. Also, in the power-minimization problem (14), the optimal objective is not the budget P but some P^*, and the connection between the two should be stated explicitly when Theorem 1 is applied to problem (5).","section":"Theorem 1, Eq. (15) and Eq. (16)"},{"comment":"The summation in the second term of the optimality condition should run over the channel index i, not the user index k; that is, the expression should be Σ_{i=1}^K h_i(s)∫_S h_i^*(z) w_k^†(z) dz. The current notation is confusing and appears to be a typo.","section":"Appendix E, Eq. (93)"},{"comment":"The text states that the Pareto property allows the optimum to be obtained 'in polynomial time' using polyblock outer approximation, but Section IV-B later states that the algorithm has prohibitive complexity growing exponentially with K. This is contradictory and should be corrected.","section":"Section IV-A, after Eq. (18)"}],"recommendation":"major_revision","confidential_remarks":"The strong-duality gap is the main obstacle; it is likely fixable by adding the standard phase-normalization/SOCP reformulation and a rigorous infinite-dimensional duality argument, or by directly proving sufficiency of the KKT conditions for the convex reformulation. If the authors cannot provide such a proof, the global optimality claims should be substantially weakened. The fixed-point convergence proof should also be added. Given the otherwise sound algebraic structure and useful numerical study, I do not recommend rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a solid, incremental paper. The closed-form optimal structure for CAPA beamforming is the right continuous analog of the classic discrete-array result, and I believe it is correct—but the proof leans on a strong-duality assertion that is handwaved in Appendix D, and that gap should be closed before publication.\n\nWhat's new: the paper gives the first global optimal transmit beamforming for continuous aperture arrays, with structure w(s)=h(s)(I+ΛQ/σ^2)^{-1}P^{1/2}, and backs it with a monotonic optimization algorithm, plus MRT/ZF/MMSE heuristics and asymptotic optimality results. The inverse-operator lemma and the calculus-of-variations step are clean; I checked the algebra and did not find errors. The MMSE/SLNR proof is a nice extra. Citation pattern is fine: self-citations to [25],[31] are to prior suboptimal CAPA methods, not circular inputs.\n\nThe soft spots are real. The stress-test note lands. Appendix D says problem (14) is convex and satisfies strong duality 'by considering h_k(s) and w_k(s) as infinitesimal vectors.' That is an analogy, not a proof. As written, the SINR constraint is not convex—for K=1 the feasible set is the complement of a slab in L2. A rigorous treatment needs a phase-normalized or conic reformulation and a Slater argument in the infinite-dimensional setting. Without it, the KKT derivation gives only necessary stationarity; the global optimality claim and the monotonic optimization machinery built on it are not fully justified. I think the result is probably right and the gap is likely fixable, but a referee should require the missing proof. Secondary, minor-to-moderate: the fixed-point iteration (27) has no convergence or uniqueness proof, and Algorithm 2 depends on it. Numerical results are informative but no code is provided; for a theory paper that is acceptable.\n\nWho this is for: people working on CAPA/holographic MIMO beamforming. It deserves a serious referee. Send it to review with a request for rigorous strong duality and a convergence analysis. I would cite it if I were working in this area.","headline":"Solid incremental step for CAPA beamforming; main result probably correct but strong-duality proof is handwaved—send to review with a demand for rigor.","tokens_in":22042,"tokens_out":4541,"would_cite":true,"duration_ms":44073,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Optimal multi-user CAPA beamforming has a closed-form matrix structure and a globally convergent fixed-point algorithm.","keywords":["continuous aperture array","CAPA","optimal beamforming","calculus of variations","monotonic optimization","polyblock outer approximation","MMSE beamforming","multi-user MIMO"],"falsifier":"Numerically solve problem (14) by very fine discretization for a small user set and compare the minimum power with $\\sum_k \\lambda_k$ from the fixed-point iteration (27); any gap between the two would show that strong duality failed for that channel and that the Theorem 1 structure is not globally optimal in that case.","tokens_in":20991,"feed_emoji":"📡","tokens_out":8884,"duration_ms":91985,"temperature":0.7,"pith_summary":"This paper is about what optimal downlink beamforming looks like when the base station's radiating surface is continuous (a CAPA) rather than a discrete antenna array. It establishes that the optimal beamformer is always a linear combination of the continuous channel responses, with coefficients fixed by the channel-correlation matrix: $w(s)=h(s)(I_K + \\sigma^{-2}\\Lambda Q)^{-1}P^{1/2}$. On that structure it builds a polyblock outer-approximation algorithm with a closed-form fixed-point iteration that reaches the globally optimal beamformer for any monotonic system utility, and it derives MRT, ZF, and MMSE designs as low-complexity special cases. The paper also proves that MRT and ZF are asymptotically optimal in the low- and high-SNR regimes, that MMSE maximizes each user's SLNR, and that simulated CAPA sum rates exceed those of equivalently sized discrete arrays.","feed_headline":"One matrix formula solves continuous-aperture beamforming","feed_subtitle":"Beamformers become weighted channel responses; CAPA beats discrete arrays, and MMSE nearly matches the optimum.","key_machinery":"The load-bearing object is the continuous integral kernel $G(s,z)=\\delta(s-z)+\\sum_{i=1}^K \\rho_i h_i(s)h_i^*(z)$ together with its closed-form inverse constructed in Lemma 2, which uses the finite matrix $D=(I_K+\\Lambda Q)^{-1}$. This 'inversion of continuous functions' identity converts the infinite-dimensional optimality condition into a $K\\times K$ matrix equation, so the continuous beamformer is fully determined by the channel-correlation matrix $Q$. Lagrangian duality supplies the multipliers $\\Lambda$, the calculus of variations supplies the stationarity condition, and the Woodbury identity reshapes the result into the compact formula of Theorem 1.","core_discovery":"The paper's central claim is Theorem 1: for a continuous aperture surface, the beamformer solving the SINR-constrained power-minimization problem, and therefore the beamformer maximizing any strictly monotonic system utility, has the closed form $w(s)=h(s)(I_K + \\sigma^{-2}\\Lambda Q)^{-1}P^{1/2}$. Here $h(s)$ stacks the continuous channel responses, $Q$ is their $K\\times K$ correlation matrix over the aperture, and $\\Lambda$ and $P$ are diagonal matrices of Lagrange multipliers and power allocations whose traces sum to the transmit power. The derivation handles inversion of the continuous kernel by a finite-matrix identity (Lemma 2), then applies the calculus of variations to the Lagrangian. Given this structure, the paper claims Algorithm 1, a polyblock outer-approximation method using bisection and a fixed-point iteration for the multipliers, returns the globally optimal CAPA beamformer for any monotonic utility. The MMSE design is then shown to be exactly SLNR-optimal under equal power allocation, and CAPA is shown numerically to outperform discrete arrays of the same aperture.","pith_inferences":["A practical consequence of the structure that the paper only hints at is that estimating the $K\\times K$ correlation matrix $Q$ may matter more than finely sampling the aperture, since $Q$ alone determines the optimal weights.","The paper's comparison holds the physical aperture fixed but not the number of degrees of freedom; a fairer follow-up would compare against a discrete array with the same number of spatial DoFs, which might narrow the reported gain.","If the strong-duality premise ever fails for a particular channel law, the same closed-form structure would still be a reasonable near-optimal heuristic, but the global-optimality guarantee in Algorithm 1 would not follow for that case."],"forward_implications":["Any strictly monotonic system utility becomes globally optimizable in principle through Algorithm 1, with the continuous part of the problem reduced to matrix arithmetic, a scalar bisection, and a fixed-point iteration.","MRT and ZF are asymptotically optimal as $\\sigma^2\\to\\infty$ and $\\sigma^2\\to0$, respectively, so those SNR regimes have a closed-form, low-complexity optimal beamformer.","MMSE is the exact SLNR-optimal design under equal power allocation and is numerically near-optimal across the tested SNR range, aperture sizes, and user counts.","Simulated CAPA sum rate exceeds that of an equally sized SPDA, for example by 30% at $P=10\\,\\mathrm{mA}^2$, with the gain attributed to beamforming gain and the multiplexing gain left unchanged.","Any beamformer component orthogonal to the users' channel responses is useless: it consumes power but leaves every SINR unchanged."],"supporting_citations":[{"why":"Supplies the SPDA optimal-beamformer structure and the convexity/strong-duality facts that the paper extends to continuous functions.","marker":"[34]"},{"why":"Provides the convex optimization result used in Appendix D to argue strong duality for the infinitesimal-vector discretization.","marker":"[41]"},{"why":"Supplies the polyblock outer-approximation algorithm and its global-optimality theory for monotonic optimization.","marker":"[38]"},{"why":"Provides the monotonic-optimization foundations and the closed-form power-allocation scheme used for the utility objectives.","marker":"[39]"},{"why":"Prior calculus-of-variations CAPA beamforming method used as the suboptimal benchmark that the global algorithm is compared against.","marker":"[31]"},{"why":"Fourier-based pattern-division multiplexing benchmark for CAPA and the consistent SPDA channel and power model used in simulations.","marker":"[29]"},{"why":"Wavenumber-division multiplexing model for CAPA that fixes the aperture, power, and channel conventions underlying the CAPA-SPDA comparison.","marker":"[24]"}],"fun_headline_variants":["Continuous-aperture beamforming solved in closed form","Optimal beamforming for continuous aperture arrays","CAPA beamforming: closed-form optimum and global algorithm","Multi-user CAPA: one formula for optimal beamforming","Continuous array beamforming: from calculus to optimality"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument assumes, without a direct proof, that the continuous power-minimization problem (14) has zero duality gap—that Lagrangian methods find the exact optimum—and if that assumption fails the closed-form beamformer need not be globally optimal.","fun_headline_variants_meta":{"raw":{"variants":["Continuous-aperture beamforming solved in closed form","Optimal beamforming for continuous aperture arrays","CAPA beamforming: closed-form optimum and global algorithm","Multi-user CAPA: one formula for optimal beamforming","Continuous array beamforming: from calculus to optimality"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000205,"raw_usage":{"total_tokens":1471,"prompt_tokens":1100,"completion_tokens":371,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":716,"completion_tokens_details":{"reasoning_tokens":295}},"tokens_in":716,"tokens_out":371,"duration_ms":5325,"temperature":1.0,"reasoning_tokens":295,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:43:31.516779+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically solve problem (14) by very fine discretization for a small user set and compare the minimum power with $\\sum_k \\lambda_k$ from the fixed-point iteration (27); any gap between the two would show that strong duality failed for that channel and that the Theorem 1 structure is not globally optimal in that case.","supporting_citations":[{"cited_title":"Optimal multiuser trans- mit beamforming: A difficult problem with a simple solution structure [lecture notes],","cited_arxiv_id":null,"evidence_quote":"Supplies the SPDA optimal-beamformer structure and the convexity/strong-duality facts that the paper extends to continuous functions."},{"cited_title":"Monotonic optimization in communication and networking systems,","cited_arxiv_id":null,"evidence_quote":"Supplies the polyblock outer-approximation algorithm and its global-optimality theory for monotonic optimization."},{"cited_title":"Optimal resource allocation in coordi- nated multi-cell systems,","cited_arxiv_id":null,"evidence_quote":"Provides the monotonic-optimization foundations and the closed-form power-allocation scheme used for the utility objectives."},{"cited_title":"Pattern-division multiplexing for multi-user continuous-aperture MIMO,","cited_arxiv_id":null,"evidence_quote":"Fourier-based pattern-division multiplexing benchmark for CAPA and the consistent SPDA channel and power model used in simulations."}],"review_version":1}